Requirement management evidence file intelligent review method based on RAG technology

Abstract

The invention relates to the technical field of demand management evidence file review, in particular to a demand management evidence file intelligent review method based on the RAG technology, which comprises the following steps: preprocessing a demand technology document, including format unification, noise removal and preliminary analysis, refining and structuring extracted demand items, and analyzing the demand technology document; and the structured demand entries are converted into inquiry statements, so that a large language model can understand and answer the inquiry statements conveniently. According to the method, entry demands are obtained by analyzing the demand technology document, quantitative management of the demands is facilitated, digital management of evidence files is realized, digital transformation of enterprises is promoted, intelligent matching of the demand management evidence files based on the RAG technology is realized, the workload of evidence file examination is reduced, and the efficiency is improved. The association degree of the evidence file and the demand is improved, and automatic tracing of the demand and the evidence file is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of demand management evidence document review, and in particular to a demand management evidence document intelligent review method based on RAG technology. Background Art

[0002] Demand management plays an important role in the research and development of rail transit products. It can ensure that products meet user needs, improve project success rates, optimize resource allocation, etc. Evidence document review is a crucial part of demand management. Evidence document review mainly examines whether the content of the evidence document can meet the technical requirements in the demand document.

[0003] However, in general, although rule-based methods have high accuracy for specific types of documents or documents with clear formats, they have poor flexibility and are difficult to adapt to changes in document formats or new types of documents. As the types and numbers of documents increase, the rule sets need to be constantly updated and improved, and the processing capabilities for complex documents are limited, making it impossible to accurately identify them. Although deep learning-based methods have the advantages of strong adaptability, high accuracy, and strong scalability, they require a large amount of labeled data and computing resources before use, and the training cost is relatively high. If there is noise or bias in the training data, it may also affect the performance of the model. At the same time, the internal working mechanism of the deep learning model is difficult to understand intuitively, resulting in difficult results to explain and debug. Therefore, the traditional manual review method of evidence documents is inefficient, and reviewers need to open a large number of evidence documents one by one to see if there is any content that can respond to the required items.

[0004] Based on this, the present invention provides a method for intelligent review of demand management evidence files based on RAG technology to solve the technical problems raised above. Summary of the invention

[0005] The purpose of the present invention is to provide a method for intelligent review of demand management evidence files based on RAG technology to solve the problems raised by the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for intelligent review of evidence documents for demand management based on RAG technology is provided, comprising the following steps:

[0008] S1. Preprocess the technical documents of the requirements, including format unification, noise removal and preliminary analysis;

[0009] S2. Refine and structure the extracted requirements to make them easier to understand and process;

[0010] S3. Convert the structured requirement items into query statements so that the large language model can understand and answer them;

[0011] S4. Parse and identify evidence documents and build them into a structured knowledge base;

[0012] S5. Introduce RAG technology to enhance the retrieval and matching capabilities of evidence documents;

[0013] S6. Use the large language model to further process the candidate answers and generate the final answer;

[0014] S7. In the process of generating answers, semantic similarity matching technology is introduced to improve the accuracy and relevance of answers;

[0015] S8. Integrate the generated responses and related evidence document segments into an evidence document response report;

[0016] S9. Intelligently optimize the review results and provide feedback on the generated evidence document response report to improve the review quality and efficiency.

[0017] Preferably, the implementation steps of step S1 are:

[0018] S1.1. Use automated tools to convert documents in different formats such as Word, PDF, and Excel into plain text format to ensure consistency in subsequent processing, and perform format verification on the converted text to ensure that there is no missing or incorrect information;

[0019] S1.2. Apply regular expression matching and semantic analysis to remove irrelevant information in the text, including advertisements, watermarks, headers and footers, and recheck the text after noise removal to ensure the integrity and accuracy of the information;

[0020] S1.3. Analyze the document structure, including chapters, paragraphs, and lists, extract the preliminary outline of the requirement items, and further refine and identify the requirement items based on their characteristics, including keywords and sentence patterns;

[0021] S1.4. Organize the extracted requirement items into a structured data format, including JSON or XML, to facilitate subsequent processing and analysis, and verify and organize the structured data to ensure data accuracy and consistency.

[0022] Preferably, the implementation steps of step S2 are:

[0023] S2.1. Use the semantic analysis function in natural language processing technology to conduct in-depth semantic understanding of requirement items, identify and extract key information in requirement items, including requirement type, priority, and related modules;

[0024] S2.2. Based on the extracted key information, construct a structured model of the requirement items, and verify and validate the structured model to ensure that it can accurately reflect the true intention of the requirement items;

[0025] S2.3. Clean and organize the data in the structured model, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data to ensure the accuracy and completeness of the data.

[0026] Preferably, the implementation steps of step S3 are:

[0027] S3.1. Use natural language generation technology to convert key information in the requirement items into natural language query statements, and perform syntax and semantic verification on the generated query statements to ensure that they comply with natural language specifications;

[0028] S3.2. Apply the question rewriting strategy to optimize the generated query statements to improve their clarity and accuracy, and recheck and verify the rewritten query statements to ensure that they can accurately reflect the true intention of the requirement items;

[0029] S3.3. In order to improve the efficiency and accuracy of retrieval, multiple query statements with similar semantics to the original query statement but different expressions can be generated, and the generated diversified query statements can be verified and sorted to ensure that they can cover all aspects of the demand items.

[0030] Preferably, the implementation steps of step S4 are:

[0031] S4.1. Use document parsing tools to convert unstructured evidence documents into structured data, including identifying the title, paragraph, and table structure elements in the document, and performing preliminary verification and organization of the parsed structured data;

[0032] S4.2. Perform in-depth cleaning and organization of the parsed structured data, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data to ensure its accuracy and completeness;

[0033] S4.3. Vectorize the cleaned structured data and convert it into a numerical form that can be efficiently processed in a computer, build an evidence document knowledge base, and store the vectorized data in a database for retrieval and matching in subsequent steps.

[0034] Preferably, the implementation steps of step S5 are:

[0035] S5.1. Using the retrieval enhancement function of RAG technology, retrieve document fragments related to the query statement in the evidence document knowledge base, perform preliminary sorting and screening on the retrieval results, and extract the most relevant document fragments as candidate answers;

[0036] S5.2. Perform semantic analysis on candidate answers, calculate the semantic similarity between them and the query statement, and further sort and screen the candidate answers based on the semantic similarity to ensure that the final answer provided is highly relevant to the query statement.

[0037] Preferably, the implementation steps of step S6 are:

[0038] S6.1. Input candidate answers into the large language model, use the model's contextual understanding ability to deeply interpret and analyze the answers, and further optimize and adjust the answers based on the contextual information;

[0039] S6.2. Use the knowledge extraction capability of the large language model to extract key information related to the query sentence from the candidate answers, and integrate the extracted key information into the final answer;

[0040] S6.3. Perform grammatical and semantic verification on the generated answers to ensure that they conform to natural language specifications and are accurate. Further optimize and adjust the verified answers to improve their readability and comprehensibility.

[0041] Preferably, the implementation steps of step S7 are:

[0042] S7.1. Calculate the semantic similarity between the generated answer and the original query statement, including further optimizing and adjusting the answer if the similarity is lower than a set threshold;

[0043] S7.2. Conduct in-depth semantic analysis on the generated answers, identify and extract key information and semantic features, and compare and analyze the extracted key information and semantic features with the original query statement to ensure the accuracy and relevance of the answers;

[0044] S7.3. Provide feedback and adjust the generated answers based on the results of the semantic similarity matching technology, and re-check and verify the adjusted answers to ensure that they meet the requirements.

[0045] Preferably, the implementation steps of step S8 are:

[0046] S8.1. Extract the generated responses and relevant evidence document fragments, and organize and format the extracted data for report generation in subsequent steps;

[0047] S8.2. Generate a document response report in accordance with the predetermined format and requirements, clearly presenting the response content, document citations, and related analysis and interpretation in the report;

[0048] S8.3. Verify and validate the generated reports to ensure they are accurate and in the required format, and further optimize and adjust the verified reports to improve their readability and understandability.

[0049] Preferably, the implementation steps of step S9 are:

[0050] S9.1. Use machine learning algorithms to perform quality assessments on the generated evidence document response reports to identify potential problems and deficiencies in the reports, including missing information, errors, or unclear statements;

[0051] S9.2. Optimize and adjust the report based on the results of the quality assessment, and recheck and verify the adjusted report to ensure its accuracy and completeness;

[0052] S9.3. Feedback the optimized report to the reviewer or relevant stakeholders, who will further confirm and approve the report to ensure that it meets the project or business needs.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention parses the demand technical documents to obtain itemized demands, facilitates the quantitative management of demands, parses and identifies evidence files, builds an evidence file library, and realizes digital management of evidence files, which is conducive to promoting the digital transformation of enterprises. At the same time, the present invention realizes intelligent matching of demand management evidence files based on RAG technology, realizes automatic review of the coverage and response of demand evidence files, reduces the workload of evidence file review, improves the correlation between evidence files and demands, and realizes automatic tracing of demands and evidence files. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the intelligent review method for demand management evidence documents based on RAG technology of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] See also Figure 1The present invention proposes a method for intelligent review of evidence files for demand management based on RAG technology, comprising the following steps:

[0058] S1. Preprocess the technical documents of the requirements, including format unification, noise removal and preliminary analysis;

[0059] S2. Refine and structure the extracted requirements to make them easier to understand and process;

[0060] S3. Convert the structured requirement items into query statements so that the large language model can understand and answer them;

[0061] S4. Parse and identify evidence documents and build them into a structured knowledge base;

[0062] S5. Introduce RAG technology to enhance the retrieval and matching capabilities of evidence documents;

[0063] S6. Use the large language model to further process the candidate answers and generate the final answer;

[0064] S7. In the process of generating answers, semantic similarity matching technology is introduced to improve the accuracy and relevance of answers;

[0065] S8. Integrate the generated responses and related evidence document segments into an evidence document response report;

[0066] S9. Intelligently optimize the review results and provide feedback on the generated evidence document response report to improve the review quality and efficiency.

[0067] In this embodiment, it should be noted that the implementation steps of step S1 are:

[0068] S1.1. Use automated tools to convert documents in different formats such as Word, PDF, and Excel into plain text format, and perform format verification on the converted text;

[0069] S1.2. Apply regular expression matching and semantic analysis to remove irrelevant information in the text, including advertisements, watermarks, headers and footers, and recheck the text after noise removal;

[0070] S1.3. Analyze the document structure, extract the preliminary outline of the requirement items, and further refine and identify the requirement items according to their characteristics;

[0071] S1.4. Organize the extracted requirement items into a structured data format to facilitate subsequent processing and analysis, and verify and organize the structured data.

[0072] In this embodiment, it should be noted that the implementation steps of step S2 are:

[0073] S2.1. Use the semantic analysis function in natural language processing technology to conduct in-depth semantic understanding of demand items, identify and extract key information in demand items;

[0074] S2.2. Construct a structured model of the requirement items based on the extracted key information, and verify and validate the structured model;

[0075] S2.3. Clean and organize the data in the structured model, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data.

[0076] In this embodiment, it should be noted that the implementation steps of step S3 are:

[0077] S3.1. Use natural language generation technology to convert key information in the requirement items into natural language query statements, and perform syntax and semantic verification on the generated query statements;

[0078] S3.2. Apply the question rewriting strategy to optimize the generated query statements to improve their clarity and accuracy, and recheck and verify the rewritten query statements;

[0079] S3.3. In order to improve the efficiency and accuracy of retrieval, multiple query statements with similar semantics to the original query statement but different expressions can be generated, and the generated diversified query statements can be verified and sorted.

[0080] In this embodiment, it should be noted that the implementation steps of step S4 are:

[0081] S4.1. Use document parsing tools to convert unstructured evidence documents into structured data, and perform preliminary verification and organization on the parsed structured data;

[0082] S4.2. Perform in-depth cleaning and organization of the parsed structured data, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data;

[0083] S4.3. Perform vectorization on the cleaned structured data, convert it into a numerical form that can be efficiently processed in a computer, build an evidence document knowledge base, and store the vectorized data in a database.

[0084] In this embodiment, it should be noted that the implementation steps of step S5 are:

[0085] S5.1. Using the retrieval enhancement function of RAG technology, retrieve document fragments related to the query statement in the evidence document knowledge base, perform preliminary sorting and screening on the retrieval results, and extract the most relevant document fragments as candidate answers;

[0086] S5.2. Perform semantic analysis on candidate answers, calculate the semantic similarity between them and the query statement, and further sort and screen the candidate answers based on the semantic similarity.

[0087] In this embodiment, it should be noted that the implementation steps of step S6 are:

[0088] S6.1. Input candidate answers into the large language model, use the model's contextual understanding ability to deeply interpret and analyze the answers, and further optimize and adjust the answers based on the contextual information;

[0089] S6.2. Use the knowledge extraction capability of the large language model to extract key information related to the query sentence from the candidate answers, and integrate the extracted key information into the final answer;

[0090] S6.3. Perform syntax and semantic verification on the generated answers, and further optimize and adjust the verified answers.

[0091] In this embodiment, it should be noted that the implementation steps of step S7 are:

[0092] S7.1. Calculate the semantic similarity between the generated answer and the original query statement, and then further optimize and adjust the answer;

[0093] S7.2. Conduct in-depth semantic analysis on the generated answers, identify and extract key information and semantic features, and compare and analyze the extracted key information and semantic features with the original query statements;

[0094] S7.3. Provide feedback and adjust the generated answers based on the results of the semantic similarity matching technology, and re-check and verify the adjusted answers.

[0095] In this embodiment, it should be noted that the implementation steps of step S8 are:

[0096] S8.1. Extract the generated responses and relevant evidence document fragments, and organize and format the extracted data;

[0097] S8.2. Generate a document response report in accordance with the predetermined format and requirements, clearly presenting the response content, document citations, and related analysis and interpretation in the report;

[0098] S8.3. Verify and validate the generated report, and further optimize and adjust the verified report.

[0099] In this embodiment, it should be noted that the implementation steps of step S9 are:

[0100] S9.1. Use machine learning algorithms to assess the quality of the generated evidence document response reports and identify potential problems and deficiencies in the reports;

[0101] S9.2. Optimize and adjust the report based on the results of the quality assessment, and recheck and verify the adjusted report;

[0102] S9.3. Feedback the optimized report to the reviewer or relevant stakeholders, who will further confirm and approve the report.

[0103] Example

[0104] In practical applications, the above-mentioned intelligent review method for demand management evidence files based on RAG technology specifically includes the following steps:

[0105] S1. Preprocess the technical documents of the requirements, including format unification, noise removal and preliminary analysis:

[0106] Specifically, the following steps are included:

[0107] S1.1. Use automated tools to convert documents in different formats such as Word, PDF, and Excel into plain text format to ensure consistency in subsequent processing, and perform format verification on the converted text to ensure that there is no missing or incorrect information;

[0108] S1.2. Apply regular expression matching and semantic analysis to remove irrelevant information in the text, including advertisements, watermarks, headers and footers, and recheck the text after noise removal to ensure the integrity and accuracy of the information;

[0109] S1.3. Analyze the document structure, including chapters, paragraphs, and lists, extract the preliminary outline of the requirement items, and further refine and identify the requirement items based on their characteristics, including keywords and sentence patterns;

[0110] S1.4. Organize the extracted requirement items into a structured data format, including JSON or XML, to facilitate subsequent processing and analysis, and verify and organize the structured data to ensure data accuracy and consistency;

[0111] S2. Refine and structure the extracted requirements to make them easier to understand and process;

[0112] Specifically, the following steps are included:

[0113] S2.1. Use the semantic analysis function in natural language processing technology to conduct in-depth semantic understanding of requirement items, identify and extract key information in requirement items, including requirement type, priority, and related modules;

[0114] S2.2. Based on the extracted key information, construct a structured model of the requirement items, and verify and validate the structured model to ensure that it can accurately reflect the true intention of the requirement items;

[0115] S2.3. Clean and organize the data in the structured model, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data to ensure the accuracy and completeness of the data;

[0116] S3. Convert the structured requirement items into query statements so that the large language model can understand and answer them;

[0117] Specifically, the following steps are included:

[0118] S3.1. Use natural language generation technology to convert key information in the requirement items into natural language query statements, and perform syntax and semantic verification on the generated query statements to ensure that they comply with natural language specifications;

[0119] S3.2. Apply the question rewriting strategy to optimize the generated query statements to improve their clarity and accuracy, and recheck and verify the rewritten query statements to ensure that they can accurately reflect the true intention of the requirement items;

[0120] S3.3. In order to improve the efficiency and accuracy of the search, multiple query statements with similar semantics to the original query statement but different expressions can be generated, and the generated diversified query statements are verified and sorted to ensure that they can cover all aspects of the demand items;

[0121] S4. Parse and identify evidence documents and build them into a structured knowledge base;

[0122] Specifically, the following steps are included:

[0123] S4.1. Use document parsing tools to convert unstructured evidence documents into structured data, including identifying the title, paragraph, and table structure elements in the document, and performing preliminary verification and organization of the parsed structured data;

[0124] S4.2. Perform in-depth cleaning and organization of the parsed structured data, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data to ensure its accuracy and completeness;

[0125] S4.3. Perform vectorization on the cleaned structured data, convert it into a numerical form that can be efficiently processed in a computer, build an evidence document knowledge base, and store the vectorized data in a database for retrieval and matching in subsequent steps;

[0126] S5. Introduce RAG technology to enhance the retrieval and matching capabilities of evidence documents;

[0127] Specifically, the following steps are included:

[0128] S5.1. Using the retrieval enhancement function of RAG technology, retrieve document fragments related to the query statement in the evidence document knowledge base, perform preliminary sorting and screening on the retrieval results, and extract the most relevant document fragments as candidate answers;

[0129] S5.2. Perform semantic analysis on candidate answers, calculate the semantic similarity between them and the query statement, and further sort and screen the candidate answers based on the semantic similarity to ensure that the final answer provided is highly relevant to the query statement;

[0130] S6. Use the large language model to further process the candidate answers and generate the final answer;

[0131] Specifically, the following steps are included:

[0132] S6.1. Input candidate answers into the large language model, use the model's contextual understanding ability to deeply interpret and analyze the answers, and further optimize and adjust the answers based on the contextual information;

[0133] S6.2. Use the knowledge extraction capability of the large language model to extract key information related to the query sentence from the candidate answers, and integrate the extracted key information into the final answer;

[0134] S6.3. Perform grammatical and semantic verification on the generated answers to ensure that they conform to natural language specifications and are accurate, and further optimize and adjust the verified answers to improve their readability and comprehensibility;

[0135] S7. In the process of generating answers, semantic similarity matching technology is introduced to improve the accuracy and relevance of answers;

[0136] Specifically, the following steps are included:

[0137] S7.1. Calculate the semantic similarity between the generated answer and the original query statement, including further optimizing and adjusting the answer if the similarity is lower than a set threshold;

[0138] S7.2. Conduct in-depth semantic analysis on the generated answers, identify and extract key information and semantic features, and compare and analyze the extracted key information and semantic features with the original query statement to ensure the accuracy and relevance of the answers;

[0139] S7.3. Provide feedback and adjust the generated answers based on the results of the semantic similarity matching technology, and recheck and verify the adjusted answers to ensure that they meet the requirements;

[0140] S8. Integrate the generated responses and related evidence document segments into an evidence document response report;

[0141] Specifically, the following steps are included:

[0142] S8.1. Extract the generated responses and relevant evidence document fragments, and organize and format the extracted data for report generation in subsequent steps;

[0143] S8.2. Generate a document response report in accordance with the predetermined format and requirements, clearly presenting the response content, document citations, and related analysis and interpretation in the report;

[0144] S8.3. Verify and validate the generated reports to ensure they are accurate and in the required format, and further optimize and adjust the verified reports to improve their readability and understandability;

[0145] S9. Optimize and provide feedback on the intelligent review results of the generated evidence document response report to improve the review quality and efficiency;

[0146] Specifically, the following steps are included:

[0147] S9.1. Use machine learning algorithms to perform quality assessments on the generated evidence document response reports to identify potential problems and deficiencies in the reports, including missing information, errors, or unclear statements;

[0148] S9.2. Optimize and adjust the report based on the results of the quality assessment, and recheck and verify the adjusted report to ensure its accuracy and completeness;

[0149] S9.3. Feedback the optimized report to the reviewer or relevant stakeholders, who will further confirm and approve the report to ensure that it meets the project or business needs.

[0150] Through the above steps, the present invention parses the demand technical documents to obtain itemized demands, facilitates the quantitative management of demands, parses and identifies evidence files, builds an evidence file library, and realizes digital management of evidence files, which is conducive to promoting the digital transformation of enterprises. At the same time, the present invention realizes intelligent matching of demand management evidence files based on RAG technology, realizes automatic review of the coverage and response of demand evidence files, reduces the workload of evidence file review, improves the correlation between evidence files and demands, and realizes automatic tracing of demands and evidence files.

[0151] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0152] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent review of demand management evidence documents based on RAG technology, characterized in that: The following steps are involved: S1. Preprocess the technical documents of the requirements, including format unification, noise removal and preliminary analysis; S2. Refine and structure the extracted requirements to make them easier to understand and process; S3. Convert the structured requirement items into query statements so that the large language model can understand and answer them; S4. Parse and identify evidence documents and build them into a structured knowledge base; S5. Introduce RAG technology to enhance the retrieval and matching capabilities of evidence documents; S6. Use the large language model to further process the candidate answers and generate the final answer; S7. In the process of generating answers, semantic similarity matching technology is introduced to improve the accuracy and relevance of answers; S8. Integrate the generated responses and related evidence document segments into an evidence document response report; S9. Intelligently optimize the review results and provide feedback on the generated evidence document response report to improve the review quality and efficiency.

2. According to the RAG technology-based intelligent review method for demand management evidence files according to claim 1, it is characterized in that: The implementation steps of step S1 are: S1.

1. Use automated tools to convert documents in different formats such as Word, PDF, and Excel into plain text format, and perform format verification on the converted text; S1.

2. Apply regular expression matching and semantic analysis to remove irrelevant information in the text, including advertisements, watermarks, headers and footers, and recheck the text after noise removal; S1.

3. Analyze the document structure, extract the preliminary outline of the requirement items, and further refine and identify the requirement items according to their characteristics; S1.

4. Organize the extracted requirement items into a structured data format to facilitate subsequent processing and analysis, and verify and organize the structured data.

3. According to claim 2, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S2 are: S2.

1. Use the semantic analysis function in natural language processing technology to conduct in-depth semantic understanding of demand items, identify and extract key information in demand items; S2.

2. Construct a structured model of the requirement items based on the extracted key information, and verify and validate the structured model; S2.

3. Clean and organize the data in the structured model, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data.

4. According to claim 3, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S3 are: S3.

1. Use natural language generation technology to convert key information in the requirement items into natural language query statements, and perform syntax and semantic verification on the generated query statements; S3.

2. Apply the question rewriting strategy to optimize the generated query statements to improve their clarity and accuracy, and recheck and verify the rewritten query statements; S3.

3. In order to improve the efficiency and accuracy of retrieval, multiple query statements with similar semantics to the original query statement but different expressions can be generated, and the generated diversified query statements can be verified and sorted.

5. According to claim 4, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S4 are: S4.

1. Use document parsing tools to convert unstructured evidence documents into structured data, and perform preliminary verification and organization on the parsed structured data; S4.

2. Perform in-depth cleaning and organization of the parsed structured data, remove duplicate, redundant or erroneous information, and recheck and verify the cleaned data; S4.

3. Perform vectorization on the cleaned structured data, convert it into a numerical form that can be efficiently processed in a computer, build an evidence document knowledge base, and store the vectorized data in a database.

6. According to claim 5, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S5 are: S5.

1. Using the retrieval enhancement function of RAG technology, retrieve document fragments related to the query statement in the evidence document knowledge base, perform preliminary sorting and screening on the retrieval results, and extract the most relevant document fragments as candidate answers; S5.

2. Perform semantic analysis on candidate answers, calculate the semantic similarity between them and the query statement, and further sort and screen the candidate answers based on the semantic similarity.

7. According to claim 6, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S6 are: S6.

1. Input candidate answers into the large language model, use the model's contextual understanding ability to deeply interpret and analyze the answers, and further optimize and adjust the answers based on the contextual information; S6.

2. Use the knowledge extraction capability of the large language model to extract key information related to the query sentence from the candidate answers, and integrate the extracted key information into the final answer; S6.

3. Perform syntax and semantic verification on the generated answers, and further optimize and adjust the verified answers.

8. According to claim 7, a method for intelligent review of demand management evidence files based on RAG technology is characterized in that: The implementation steps of step S7 are: S7.

1. Calculate the semantic similarity between the generated answer and the original query statement, and then further optimize and adjust the answer; S7.

2. Conduct in-depth semantic analysis on the generated answers, identify and extract key information and semantic features, and compare and analyze the extracted key information and semantic features with the original query statement; S7.

3. Provide feedback and adjust the generated answers based on the results of the semantic similarity matching technology, and re-check and verify the adjusted answers.

9. The intelligent review method for demand management evidence files based on RAG technology according to claim 8 is characterized in that: The implementation steps of step S8 are: S8.

1. Extract the generated responses and relevant evidence document fragments, and organize and format the extracted data; S8.

2. Generate a document response report in accordance with the predetermined format and requirements, clearly presenting the response content, document citations, and related analysis and interpretation in the report; S8.

3. Verify and validate the generated report, and further optimize and adjust the verified report.

10. The method for intelligent review of demand management evidence files based on RAG technology according to claim 9 is characterized in that: The implementation steps of step S9 are: S9.

1. Use machine learning algorithms to assess the quality of the generated evidence document response reports and identify potential problems and deficiencies in the reports; S9.

2. Optimize and adjust the report based on the results of the quality assessment, and recheck and verify the adjusted report; S9.

3. Feedback the optimized report to the reviewer or relevant stakeholders, who will further confirm and approve the report.

Images